SEO

Why Google Uses Machine Learning – Here’s Why #245

Eric Enge and Jessica Peck on Why Google Uses Machine Learning

Machine learning is always a hot topic in the search community, and there is still confusion on how and why Google uses machine learning in their search algorithms.

In this episode of the award-winning Here’s Why digital marketing video series, Eric Enge and Jessica Peck explain why you need to understand how Google uses machine learning.

Don’t miss a single episode of Here’s Why. Click the subscribe button below to be notified via email each time a new video is published.

Subscribe to Here’s Why

Resources

Transcript

Eric: Hey, everyone. Today, I have Jessica Peck with me. Jess is a Marketing Technology Associate here at Perficient offices in Framingham. She works on our internal marketing tools as well as tactical SEO for websites. Say hi, Jess.

Jessica: Hi, everyone.

Eric: Awesome. So, we talk a lot about machine learning here at Perficient, how we use it for ourselves, how our vendors use it for SEO, and how Google uses it.

Jessica: Yes. It’s a massive topic, and even if you just talk about how Google uses it, there are tons of variations for translation, image recognition, voice, and natural language processing.

Eric: And then there’s the development of TensorFlow and other libraries and developments from the Google Brain team, too.

Jessica: Yeah. So, obviously, Google loves machine learning. And it’s useful to use machine learning for search.

Eric: We’ve talked about this before on “Here’s Why,” but how does Google use machine learning to understand the user’s query intent and to find the best content to match that intent?

Jessica: It’s been a shift over time from an exact match kind of rudimentary search system with other computational methods to give the appearance of machine understanding — like page rank, and moving towards more actual machine learning methods and trying to match intents to concepts and vice versa.

Eric: Which is a lot more complicated than just matching two strings.

Jessica: Yes. And Google wants to remove as much human interference from search results as possible. They want to serve the user the fastest, most comprehensive answer possible for the exact query and intent. And machine learning can help them do that at scale.

Eric: So, Google uses machine learning systems to find sites that have deep content experiences on a query intent level and an entity level. This is where you can see the elements of how we understand voice and visual search creep into the traditional search context, too. And Google is trying to understand queries more conversationally and building responses based on all of the elements of what an entity is, including visual signals.

Jessica: For sure. If you look at how to build a Google Assistant app, for example, the breakdown of how you build queries for that — entity, intent, etcetera — mirrors some of the ways we can expect Google to do the same thing.

Eric: So, with such a big machine learning universe around search, what should marketers be thinking about and be on the lookout for as well?

Digital Marketing - The Digital Essentials, Part 1
The Digital Essentials, Part 1

A compelling digital strategy aligns customer experiences, business execution, and the right technology as the market and your competition constantly evolve. Our Digital Essentials highlight the most compelling aspects of all three to help you react and respond to this ongoing evolution.

Get the Guide

Jessica: Well, we don’t really know all of the ways Google uses machine learning and search. And they’re certainly experimenting with new methods all the time. But for natural language understanding, there are three named important algorithms that we can talk about explicitly. The first is RankBrain.

Eric: They called it one of the top three most important ranking factors when it was released in 2015.

Jessica: Definitely. Gary Illyes has described RankBrain as the machine learning ranking component that uses historical search data to predict what a user would most likely click on for a previously unseen query. If RankBrain sees a word it isn’t familiar with, it can guess what might have a similar meaning and filter search results based on that. It attempts to map new queries into entities that have the best chance of being related to or matching it.

Eric: So, it’s great for dealing with new search queries and long-tail keyword scenarios.

Jessica: Right. It greatly helps Google to hit the ground running on new trends and entities. It’s also part of understanding user intent and language. RankBrain helps Google understand stop words like “and” and “the” when they’re important, like for the differences between the query “office” and the query “The Office.”

Eric: Absolutely. So, it’s a user intent algorithm. It can tell if a query is a news-oriented query better than traditional Google search algorithms and helps Google decide what parts of the ranking algos get applied to different searches. It picks the best existing Google algorithms to match that intent and deliver a search result.

Jessica: Exactly. Now, you can’t really optimize for RankBrain. It’s more about making sure your content is in the right place at the right time about the right things.

Eric: So, what’s the second named algorithm?

Jessica: Well, we’ve been over this before on Here’s Why. Let’s go over BERT again, and I promise not to make any more jokes about Ernie this time.

Eric: Excellent. I’ll hold you to that.

Jessica: So, BERT is a bidirectional algorithm. It greatly helps Google understand the context around individual words both forwards and backward, not just one by one.

Eric: In their post about the launch of BERT, Google said that BERT would help search better understand one in ten searches in the U.S. in English, and they’ve been expanding BERT’s capabilities to other languages since then.

Jessica: Yes. For both BERT and RankBrain, it seems like around 10% of searches get affected.

Eric: Well, for a lot of shorter tail searches, Google probably doesn’t have to try using machine learning. There are plenty of factors that go toward understanding page quality for more frequently searched terms. But BERT seems to be very useful for longer long-tail queries — I think you could say the same about RankBrain — that are less frequently searched for.

Jessica: Right. So, the examples that Google gives of searches that BERT helps with involve relationships between words that used to be kind of guessed at and are now explicit — sentences like, “Brazil traveler to USA needs a Visa” or “Can you get medicine for someone pharmacy?” Dawn Anderson, @dawnieando on Twitter, has done a lot of work on this topic, and I definitely recommend checking out her Twitter to learn more about this kind of algorithm.

Eric: So, these are two major developments in natural language processing and machine learning for Google. What’s the third?

Jessica: Neural matching. This was brought up by Danny Sullivan in 2018, and I think it’s criminally under-discussed in search circles.

Eric: So, what is neural matching?

Jessica: It’s super synonyms, matching words to concepts. So, you know when you’re trying to describe a thing, and it’s big and it’s gray and it has a trunk and it has ears and…

Eric: It’s an elephant!

Jessica: Yes! So, we did some neural matching right there with the neural networks in our heads. Neural matching is helping Google get better at understanding people when they aren’t super coherent or aren’t sure of the precise words that will get them what they want. Who hasn’t Googled something like, “songs that go dun-dun-dun” and been surprised at how close the results get?

Eric: So, how can we optimize for neural matching then?

Jessica: Well, once again, it’s not that easy. It’s less about optimizing for neural matching and more about recognizing the patterns that neural matching is an element of. How are your customers actually searching for your products or pages? Are they searching for SEO, or are they searching for, “make my site be higher in Google, please?”

Eric: It comes back to ensuring you have comprehensive, comprehendible content — content that can cover the different ways people ask questions and the detailed questions they ask, and get them to the answers they’re really looking for.

Jessica: So, write good content?

Eric: Yes. Google wants to be able to tell what is good content without having a human involved. But humans will always inevitably read your content. So, ensuring it’s good is important at every level.

Don’t miss a single episode of Here’s Why. Click the subscribe button below to be notified via email each time a new video is published.

Subscribe to Here’s Why

See all of our Here’s Why Videos | Subscribe to our YouTube Channel

About the Author

Eric Enge leads the Digital Marketing practice for Perficient. He designs studies and produces industry-related research to help prove, debunk, or evolve assumptions about digital marketing practices and their value. Eric is a writer, blogger, researcher, teacher, and keynote speaker and panelist at major industry conferences. Partnering with several other experts, Eric served as the lead author of The Art of SEO.

More from this Author

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Subscribe to the Weekly Blog Digest:

Sign Up